OpenCode Review: Because I Hit Claude Code Usage Limits (Claude Code vs OpenCode vs Codex CLI)

Key Takeaways

Yesterday I was working with Claude code on an Apple Watch fishing game. Yes, I have an Apple Watch now and I wanted to create a little game for it in which you can fish by just doing a gesture and then reeling in with with with the scroll thingy on the Apple Watch. Um like who doesn't want to fish while doing a number two in the bathroom? But anyways, that's a topic for a different video. This um what I was doing yesterday was coding a bit, but after only like 30 minutes with Opus 4.5, I didn't have any usage left. And as manual coding is like an ancient relic from 2024, I just booted up my ultra-reliable Codex, but after using Claude code so much during the holidays, I started to strongly feel like Codex is missing something. And although GPT-5.2 Codex is an extremely powerful model, there were things such as plan mode, LSPs, the Vim mode, the sub agents, and some other things that simply aren't included in the intentionally very simple CLI tool of Codex. But it seemed like God that finally noticed my struggles as in one of my daily doom scrolling sessions on X, I saw that Open Code now officially supports logging in with your OpenAI subscription. And this game in response to Anthropic basically blocking the usage of your Claude Code subscription with external services. So OpenAI basically swooped in and said, "Well, we I'll fully allow you to use your OpenAI subscription with Open Code." Which is of course very handy for them so they don't have to go all in on their own Codex CLI tool. But it's also very handy for us because from a colleague I had already heard that Open Code was quite good, but I always thought that you basically had to link an API key and me being AI fanatic or you could also call it addict. Um for me linking an API key and getting limited usage as much as my bank account provides isn't that smart of an idea. So now that So now that I can finally just use my OpenAI subscription and use my Codex usage from that, I thought it was the ideal time to check out what Open Code basically offers if it's any good and also if it's a replacement for the official Codex CLI from OpenAI. So let's basically start with um checking out the website of Open Code so I can show you all the functionality that's being offered because for the last few days I have been coding in Open Code exclusively. And [snorts] like it say like I said, it's basically an open source AI coding agent. Um I think Codex and Claude Code are also open source, but um this one is as well. And um if you look at documentation, I think the first difference is basically that you can use that you may basically choose your model provider while Codex CLI is of course coupled to OpenAI and Claude Code to Anthropic. Here you can even link things like Amazon Bedrock, Google Vertex AI, Claude, even local things like Llama C++, and basically use all these um other models um to power your coding. I think um Open Code also offers something themselves. I thought it was called um Zen. Let me search for it. Zen providers. Currently in beta. And um yeah, you can basically log in on Open Code Zen. And what it allows you to do is also not sign up with a ordered party, but through Open Code Zen you can use a variety of models. Um You also get some free ones like Big Biggle. That's a private model probably going to be released soon from one of the major parties. But also some open source Chinese models. And for the other models you basically pay um pay for input and output tokens like usual. I think they claim they can get some better um so so some better um costs if a lot of people sign up for Zen, but that might be some marketing because I don't I think these are like the normal rates which you would also pay while um signing up for Anthropic in this case directly. But anyways, um so that's one of the major differences. You can really use a lot of money lot of money. Yeah, you can also pay a lot of money, but you can also use a lot of models. Um but installation is also um quite easy like um I just to brew install and then um to step Open Code, but I already have it installed so I don't think it will do much. Maybe it will just um update it. Um of course on on Windows etc. you can um Okay, a new version of Open Code was available so it does good that I upgraded. But you can see there are a lot of um ways to install it. Um let's check out the the features. First of you have tools and this is um like the built-in like like the main tools that the coding agent itself uses to manipulate your code base. Um for example of course your AI um agent can execute some bash commands, some edit, write, read. Like these are all very normal, but um the thing that for example at Open that that Codex that Claude Code and Codex CLI don't offer is basically to really um define permissions specifically for certain built-in commands. And you can basically even specify for um certain commands if they are um allowed or not. What's also cool in that that I didn't see yet with other AI coding agents is that you can basically define your own um tools. In other um like in Codex and Claude Code, you basically had to write a skill and then in the skill description link to a Python script such that um yeah, the coding agent can invoke this tool. But it seems like with Open Code you can really like write these new built-in tools and place them here, which is of course um very advanced and I wish like also other um coding agents offered. Although I don't quite know when I would define a skill yet or just use a tool. Of course, you can also define rules for your coding agent. This is basically equivalent to the agents.markdown and the cloud.markdown. Um This will basically always be included in your system prompt and in it you can write some instructions um what the project is about, the structure, how it basically should code things, what it needs to run after um after coding a certain thing. It's a bit annoying though and I've already said this before, but um Claude uses a different format. It says cloud.markdown um to include these instructions. So when you're switching from Claude Code to Open Code, you will have to rename that to agents.markdown and I basically wish that it would just be supported and it could read both. Another useful feature of course are these different agents which you can define. Agents have their own context, their own prompt. Um you can basically uh even configure the tools that they can use, um which model they use. It's very advanced and also like the um plan mode which is offered with Open Code and is I think one of the main benefits why I would use Open Code over Codex CLI is basically an agent in itself in itself with just some different um permissions. Uh Codex CLI doesn't offer agents yet, but um Claude Code does, but it's good to see that Open Code also offers it. Of course, models um we are just going to only use the OpenAI models, but you can basically um also configure these and there are a lot of providers to choose from. There are also some themes which is of course very cool. I'll show that later, but there are a lot of themes that you can you can even define your custom themes such that the um of course the terminal that the coding agent terminal doesn't clash with your the theme of the terminal itself. You can even define key binds for all the actions of the CLI tool, which I think isn't something that the other coding agents offer. >> [snorts] >> You can also define your own commands which are basically these um the these slash commands um like they are called in the other coding agents. Of course, you can also define them here yourself. These are basically for some repeated prompts that you want um the model to invoke by just typing slash. It gives a good example here of running um the test suite and then the instruction to fix any failing tests. There are also some things like formatters um which aren't included in the other CLI tools. I also don't really know if this really is something I would want the AI coding agent to do because like you you should just invoke my pre-commit hooks or just like the formatting commands um directly instead of building it in with the CLI tool, which seems like a weird conflict of interest to me. Um [snorts] of course, you can also define some um permissions on which files um certain tools are allowed. You also have LSP servers which are basically give you give the agent some more knowledge about the language itself. Um You probably already use these when coding yourself in in things like VS Code and it basically gives you some um some error some extra error detection, but also some navigational um functionality for example going to the definition of certain functions. Of course, also MCP um is offered. Um I don't use MCP myself that often. I much prefer to use um the agent skills or even the tools here might seem useful. ACP is something that allows you your agent to work smartly together with your editor. Um, I haven't used it myself um, or at least I think I I didn't use it yet um, because it only is supported with certain editors like set and the jet brains IDEs I suppose so I haven't um, used it yet. Then the agent skills which um, allows the agent to discover new instructions. For example, I have an agent skill to build my Apple Watch project and uh, the benefit of this skill is that only a small description is loaded in the context window instead of at the entire skill and the entire script. Um, something useful is that um, even that my uh, skills are written for cloud codes um, open code is still able to discover them even though the names are literally dot cloud. So it is very cool and um, these are the custom tools which I've already shown you. So from a first sight like this is very in-depth and very advanced like to harness everything that's included is everything you would expect from a coding agent and is definitely already much more advanced than something you would find with code CLI. So yeah, let's show you what the open code is all about. Let's maybe go to my project then um, open code after you install it you can just open it by typing open code. Which again takes a bit to spin up. Um, once again because I think it's written in in TypeScript and not in rust like um, code CLI. But um, if you haven't connected to open AI yet, you can basically do um, where was it? Connect and then you can select um, open AI. And then you just basically say um, chat GPT Pro plus. Uh, then you have to paste this in your browser, you get redirected to a login window and you have signed in. So it's very um, easy to do. Another thing you can already see is like that um, the entire uh, UI seems a bit more advanced or um, organized than just using um, than than just using open AI codex. As you can see open AI codex just is very very simple. So um, maybe let's cycle through through some other commands like everything you expect is here. Um, something that's very advanced is like a lot of different um, themes are offered so you can really fine-tune open codes to um, your terminal theme. I'm using Dracula so let's um, keep on using that. With the tab button you basically switch between your agents. Currently um, I am only using the default agent the plan agent which is this real read-only mode that you've probably seen in cloud code which is used for um, first brainstorming and coming up with a feature and a good plan and afterwards with the tab button you so you switch to build mode to actually build the project. Something else that's cool and isn't offered um, with with code CLI is basically editor. By typing editor um, doesn't work because I haven't specified an editor. So you need to export an editor and far and then you go back into it and when I now type editor I can basically type my prompt in my preferred editor so I can type um, um, tell me what this project is about. And I can just um, close it and manipulate my um, my yeah, basically I can manipulate the prompt that way so I really have a nice editor to work around. It's a bit annoying that after you've written something like you don't have the a vim mode entirely integrated just like with cloud code but this is a begin. So then I will just switch to plan mode. Um, the UI um, I think is a bit more polished than um, um, than than um, than code CLI but I prefer actually still prefer cloud code so um, give me a more detailed tour. Um, I don't know why but um, for some reason I I I do think it looks nice but it could look a bit a bit nicer. I don't know. But yeah, this is just a normal coding agent as usual. First plan then build. Something else that's pretty cool is like I said I for example um, here in my project have this um, cloud codes have to have the skill to basically build my um, Apple Watch project and this is actually automatically discovered by um, by open code so when I type um, can you build my project? It will say skill command build project. It will correctly execute the the build script and normally I should get a successful build. So um, you don't need to rename all your skills. Also a normal thing like this is maybe something weird but um, it glitches a bit when you exit open code like this there are some um, you have to keep saying open code so you have to clear your um, screen but that's only minor. But of course you can also um, um, check out your previous sessions. Okay, it was um, open codes um, session list. You can see all your uh, previous sessions and um, if you type continue or you if you type session and then the session ID you can continue that session or if you type continue you basically continue your um, latest session so that's very cool. Maybe to really give you an um, example let's plan a new feature and basically show um, codex and open codes side by side to see how they um, approach it. Like um, it's just something I copy pasted um, to um, yeah, to implement a new fishing feature to actually add multiple fish. Um, I'm going to add a little instruction to it um, that says um, don't implement anything yet. This isn't necessary for open um, and just plan. This isn't necessary for um, open code of course but for um, for for codex this is necessary so I just want to give them the same um, prompt. So let's see what both um, what both do. Let's enter plan mode. Um, and you will see it like um, the the UI is maybe a bit different. Um, it seems like codex is done very fast in this case. However, it does seem to give a a really um, yeah, extensive plan. Let's see what's happening with um, open codes. And at this mo at this point uh, open code is really planning a lot while um, code CLI just gives um, a a straight-up answer while yesterday um, I gave this the exact same prompt and it was actually um, quite different with um, open codes really like um, giving a plan um, really quick. Like sometimes I noticed um, yesterday but it isn't doing it right now that um, open code did have some errors with opening some files and I think this was because um, yeah, the model really isn't trained for the tools that open code provides but in this case um, it does everything um, correctly so that's nice. But just to show you the difference um, if I do it with cloud code um, you will see I also have um, a a plan mode on. Let's let's ask it the same prompt. And let's see what um, cloud code comes up with. First of you see like um, codex CLI didn't do this but um, it first explores the code base and now also and this is I think one of the things I like the most also about cloud code is like it will ask these little questions which I can just um, it also gives some predefined um, answers and some recommendations so it asked me about different aspects. So um, do you want weather and time of day sim systems implemented implement that um, and it really thinks with you on on exactly what you want and it's much more um, interactive and for some reason um, asking these questions and giving me this menu um, open code sometimes does this and sometimes it doesn't because um, let me see. I have a video right here where um, showed the video evidence that indeed in open code like here, you can see it also um asked me these questions, but it doesn't always do this and these questions aren't always um extensive. So, just the um the quality of the plan mode of of Claude Code simply um yeah, Open Code isn't as good at that as um Claude Code, but um if you compare it to just writing the prompt, "Please don't implement anything and help me plan." um which you have to do with Code X CLI, well, then Open Code is a step up. So, I'm just going to um basically quit the plan mode also with Open Code. Um Let's maybe go to some dummy project. Um Of course, um I don't know if I need to show you this, but you can of course just ask it to code as well to implement a Flappy Bird clone. Not that I'm going to entirely show you um but this is just like if you would just use um Code X CLI directly, it will look at your code base, it will start updating files, creating to-do's. Um Uh you can see it it modifies my pi project but tunnel because it wants to include pygame. Let's say uh you can run UV sync. Let's see if it's it can do it. Um The thing is a bit um the permissions of of um of Open Code do do seem a bit more free than what you by default get with Code X CLI because um I've never explicitly given it permission to run the UV command and with Claude Code and OpenAI Codex, you always have to give permission while the default mode of Open Code is very lenient. So, you have to check out for that. Um Okay, is is everything finished? I'm actually wondering um if this actually will work. Um UV run Python main.py. >> [snorts] >> If if we really get a Flappy Bird in uh not even a minute. Okay, and we indeed have a sample Flappy Bird clone in Okay, the scoring clearly isn't working. It just resets between 0 and 1, but it was only like 1 minute um of work, which also really shows you the power of these coding models. Anyways, in conclusion, the plan mode of Claude Code is still the goat and if you have uh if you're a successful business owner with a lot of money to spend, um go ahead and spend $200 on the Claude Code um uh tier and you don't miss anything by like Open Code doesn't offer anything like that I would want to switch from Claude Code from. But um if you basically want to have a beefed-up version of Code X CLI, um you see that Code X CLI is missing a lot of features and you want to use things like LSPs and the Vim mode and the plan mode and the fine level of agents, well, then definitely check out um Open Code. Open Code by itself um it also offers some more interesting things like it can use all kinds of different um open source models or even models from like these um Chinese providers, but I personally don't really have a use for that, so that's not the main attraction for me. However, I'm definitely going to keep on using it seeing if Open Code really if the Code X CLI if the GPT-5.2 Codex model really is as good in Open Code than in um Code X CLI, but for now I have no issues at all. So, I hope you enjoyed the video. Definitely let me know what you think about Open Code. Do you think this is a good replacement for Code X CLI or even Claude Code? Um I hope you enjoyed the video and I see you in the next one.